Knowledge Graph Verification Using Reproduced Graph Differences

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing methods for automatically extracting knowledge from documents struggle to determine the accuracy and usefulness of extracted information, particularly in cases where the data does not match existing business rules, as seen in research results from patents and papers.

Innovation Solution

A computer system that includes a calculation device and storage device, which generates a knowledge graph from documents, uses a model to reproduce the extracted knowledge, and calculates a difference degree between the extracted and reproduced graphs to verify the accuracy and usefulness of the extracted knowledge.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If rule-based or machine learning methods are used to automatically extract knowledge from documents, then extraction efficiency is improved, but accuracy and usefulness determination becomes difficult

Engineering Contradiction:
Improveextraction efficiencyVSAvoidaccuracy determination
Core Design Contradiction:
ProductivityVSMeasurement precision

Solution Approach 1:

The system uses a pre-trained model to generate reproduced knowledge graphs from extracted knowledge graphs, then compares the two to provide feedback on extraction accuracy. The difference degree calculation provides quantitative feedback that helps determine whether extracted knowledge is accurate and useful, resolving the contradiction between automated extraction efficiency and accuracy determination.

Inventive Principle:
Principle #23Feedback

2Measurement precision

If existing verification methods using business data are applied, then extraction accuracy can be verified, but they cannot handle novel research results that do not match existing data

Engineering Contradiction:
Improveextraction accuracy verificationVSAvoidhandling of novel research
Core Design Contradiction:
Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The system performs preliminary action by pre-training a model on existing knowledge graphs before verification. This pre-trained model can then handle novel research results by comparing extracted knowledge against the pre-learned patterns, rather than requiring exact matches with existing business data. This allows the system to verify accuracy while adapting to novel research.

Inventive Principle:
Principle #10Preliminary action

3Measurement precision

If manual reading and extraction of knowledge from documents is performed, then accuracy can be ensured, but time consumption increases significantly

Engineering Contradiction:
Improveextraction accuracyVSAvoidtime consumption
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The system introduces an intermediary verification mechanism using difference degree calculation between extracted knowledge graphs and reproduced knowledge graphs. This intermediary approach provides automated accuracy verification without requiring manual reading, significantly reducing time consumption while maintaining extraction accuracy through the model-based comparison.

Inventive Principle:
Principle #24Intermediary (Mediator)

Data Source

PatentEP4660893A1Computer and knowledge extraction method
Publication Date: 2025.12.10 HITACHI LTD
  • EP4660893A1 patent drawingFigure 1
  • EP4660893A1 patent drawingFigure 2
  • EP4660893A1 patent drawingFigure 3

AI summary

The computer receives as an input a knowledge graph representing knowledge extracted from a document including a character string and an image and stores a model for reproducing the knowledge graph. The computer extracts, upon receiving a new document, knowledge from the new document, generates the knowledge graph of the extracted knowledge, acquires a reproduced knowledge graph by inputting the extracted knowledge graph to the model, calculates, based on a difference between the extracted knowledge graph and the reproduced knowledge graph, a difference degree representing a degree of difference of the extracted knowledge from the knowledge corresponding to the knowledge graph used for generating the model, and outputs verification information including the difference degree.